Understanding the charging and balancing process |  | |
Now that we have a large number of charger users, lets explore in more detail what happens during a charge, and how plotting the charge data can tell us about the condition of the pack.
This graph is the first full discharge of a Prius Pack that was charged several times, but mostly just sat for over 2 years.
We can see that the bottom trace(slope) took a while to get initialized, but then showed the subtle changes in the rate of discharge, as a higher or lower slope value,The voltage dropped quickly at first,which caused the slope to shoot up then settled to a stable value and began to take longer and longer to change one step in voltage.
Then during the flattest section of the discharge, we see a steady 35 as the slope during the majority of the discharge.
At the end of the discharge,the voltage begins to drop as the cells started fading, we see the rate of drop off rapidly increase to shut down the discharge.
The pack was amazingly well balanced, and may be showing us that we should examine the Prius cells as replacements for the cylindrical ones, and may have better luck?
whole pack discharger
|
Waynetc pack
|  | | | Balanced but weak pack looks good at 2 A discharger rate |
Waynetc from Insight Central sent his pack to me via Ray Holan who was coming out for a charger harness and MIMA install.
Wayne had discharged the pack so we started with an 18 hour soak charge, and then discharged. I did not have the datalogger available as Ray was setting up a Database program for me to give me better order tracking and inventory control. (Thanks Ray)
The pack discharged down to the Minimum discharge voltage I had set up at 135V. Got 130 minutes of discharge.As we can see from the close up, and the slope graph not showing the charistic increase in discharge slope that we have seen before, so this pack probably could have been discharged to a lower voltage, the slope ending on the flat shows it still has more to go before depleted.
I recharged and set up the datalogger to capture the second cycle, which ran for 166 minutes of discharge. We can see from the nice reasonably flat discharge slope graph that the discharge proceeded at a reasonably nice rate of 25. Lower the number the better the pack can provide current under the ~2A load. Have seen 4-5 minimum slope on really good pack, and as high as 120 slope on the really bad silver pack with multiple cell drop offs.
Since 2 A is a low discharge compared to the cars 100A, we will need to do more testing to see how the 2A data reflects the higher currents in the car, when I get the 60A stick discharger ready.
Wayne reports that his car would only provide 2-3 seconds of assist and then it drops to 33 - 50% assist - this is with or without mima. He can do this over and over again, or hold the joystick forward and get very mild boost. Then after a while the assist gauge will go from 19 - 20 bars to about 3 and it starts to charge up again. That's when it pops codes - usually 1447 - 1149.
High IR?
We will see what happens.
He will let the pack sit for a week while he does other work, and will do another discharge then, so he can get an idea of what the self discharge is like.
Will log the self discharge over the week.
We are looking at making an automatic stick by stick 60A discharge tester that would tie into the discharger jack.
A new mode 8 will be made (Code V2.3). The idea would be to open the pack. clip big alligator clips to the ends of each stick,one at a time.
A press of the start button,would do a test which would read the whole pack voltage, turn on the 60A load, wait a programmable delay, take another reading, then turn off the load. The difference between the two voltages will represent the high current discharge internal resistance of just that stick.
Run through the pack,write down the results and we will have an accurate stick by stick internal resistance value.
Lets see how Waynes pack works regarding self discharge as well as how it responds to the real world load of normal driving after he gets it back in the car, and may want to use his pack for the high current stick level test to help us better understand how to qualify the packs and the subpacks within it.
|
More silver pack testing
|  | | | Second test of Silver pack (bad one) |
Did a workaround for my Labview problems, so I could continue to do some testing.
The graphs above are the same one as below, but this time I traced it with and without the drop out detect working. We see the same 5 cells drop out.
The drop out detect is looking for the rate of change signature of a drop out in the data.
Some close ups show that the dual running average difference is a good indicator of the rate of change, and works well for detecting too rapid drop off, topping detection on the charge side should work more reliably that the sample timer system. |
Getting past the drop out to see what is happening.
|  | | | Seeing what comes after a dropout and testing the slope detect |
Charged the pack very well overnight, and did another discharge,with the automatic detection only giving warnings instead of stopping,so I can discharge as deeply as I need to during this important test.This time I display the long term average of the last 50 datapoints,(Yellow) and the short term average (red) on the second graph.
On the third graph, I show the difference between the two averages. The ripple is the effect on the short term average as a bit change happens.
We see that the actual slope of discharge relates exactly to this difference. When first starting the discharge after a charge, we see a big number(missed it on this graph),but it was near 100 this number drops as the slope flattens out, and gets into the stable zone. On this Civic pack, on the top photo, you can see how consistent the rate of change was through most of the discharge.
The differential between the two averages was at 6-8 through much of this linear slope. The cell drop out, and the 4 others that followed were at a bit lower voltage than the drop out yesterday, so the cell is still improving.
The blip in the slope detector trace shows the rise in drop out slope when the first cell (subpack 10?dropped out,and that raised the difference from the 6-8 to about 25, and a more substantial increase when the rest started to drop, which reached over 140 before I stopped it manually.Clearly a nice clean signal.
The first slope change triggered the time/bit change detection system,and would have gone back to charge.
Once past the detection point, the charger was beeping at me.
I need to know if it was a real drop out, so I had to let it go. I figured I needed to see if any others were ready to go, and sure enough and saw 4 more cells take a dive before I stopped.
I used this opportunity to remove the ends of the pack, and try to identify the weakest subpacks.
I turned on the discharger and recorded the following voltages.
1) 5.60 *
2) 5.72 *
3) 7.03
4) 6.92
5) 5.59 *
6) 5.77 *
7) 7.04
8) 7.09
9) 6.99
10) 4.01 *
11) 6.95
12) 6.94
13) 7.07
14) 6.96
15) 6.98
16) 7.06
17) 7.05
18) 6.99
19) 7.05
20) 7.04
Looks like sticks 1,2,5,6, lost a cell, and stick 10 lost 2.
At this point I could pull those sticks, and determine the bad cells, replace them, and the pack would be in much better shape.
May present a way to fix a pack fairley cheaply?
I will get my cell level tester going again, and see what the cells in the weaker sticks look like.
|
The signature of a cell drop out
|  | | | whole drop out event |
For testing the pack discharger, we need to get some data on what a typical cell dropout looks like, so we can tune the dropout detection system.
This test version of the charger software allows us to hear a beep when the detection system wants to stop the discharge and start the recharge, but allows us to let it proceed past that point to be sure the detection caught a real drop out.
Here we see the cell drop out in three views, 1 as a part of the larger graph, 2, zoomed in with only the cell drop out filling the screen, and 3, using excel to graph the recorded data.
We see just about 1 V of drop at a much faster rate than the trend. One bit / 5 second samp-le, or about 20-30 seconds for the full drop. The discharge slope after the drop out went back to the same slope as before the drop out.
The detection system picked upthe drop out a bit after the second sample. Worked nicely.
You can see from this graph, that this 20-30 second event could have easily been missed just by looking at the display, as there is no trend line.
The Labview screen shot is part of the revised program taking into account some of the lessons learned last Sundays Labview class |
The Big picture
The car has taps at each 12 cell section of the pack, and we have been assuming all these years that it is specifically for watching for a drop out as well as looking at the tap to tap match. that is the drop out protection that is doing the recal activation, which we also assume is the recovery from a drop out. This means we expect the car to stop assist when a cell drops out, so if our assumptions are correct, which I have not seen any evidence to the contrary, using the car to cycle while not as deep a discharge, is totally safe.
The detection of a cell drop out in the middle of the full 120 cells is 10 times more difficult, and would never work in the car, as we are constantly charging and discharging. With a discharger under the chargers control, we have a steady state ~2A discharge, and the voltage drop over time is a good indicator of a drop out mainly because it is so steady, and when the cell drops out, we see a quick drop of ~0.6V to 0.8V, then the slope returns to the earlier slope. On a fully balanced pack, we will still see the weakest cell drop out first, but it could wait until the whole pack is starting its dive.The lower the drop out voltage the better the pack.
If a load is applied, the voltage immediately drops due to the IR, then it drops due to depleting charge. The more current the larger the voltage shift. On the charge side we again immediately see a voltage rise, that is proportional to the current.
A pack will increase in IR as the cells age.
A voltmeter, or Peters OBDII gauge, will show the battery voltage under the various real time loads and charge conditions, and a good approximation of pack overall health, is possible by watching how low the voltage drops under full assist conditions. A resting (no assist or regen) 155V pack, may drop to less than 120V during heavy assist, and then pop right back to nearly the same 155V on returning to the resting state. A better pack will not drop as far with the same load. Same on the regen. If the pack hits 180V+ during full regen,and it drops very low during assist, it is suffering from IR issues, and may not recover as well with cycling as a pack that has lower IR.
I believe the OBDII gauge has an IR measurement based on this voltage drop, but the measurement suffers from the delayed recovery of the voltage after a shot of assist or regen. The voltage stays elevated for a period of time after a burst of charge, and stays depressed after a burst of assist, so the voltage is bouncing all over the place, which makes this a difficult measurement to make with high accuracy, but a relative IR reading is still very useful in answering the big " how do I know the condition of my pack" question
A very slow discharge may not show a drop out as easily as a higher current faster one, as the difference between the rest of the pack discharge slope could be quite similar to the cell drop out slope and be difficult to pick up.
I begin the testing of our new drop out detect software this morning.
We now have several detection processes operating in parallel.
1. If the pack is dropping too quickly during the initial slope detect which starts at a voltage we set for the pack voltage, it indicates a very weak pack.
2. We compute an average slope based on the time it takes for the pack volts to drop 5 bits. this sets the detection target rate of drop, which works quite well on any fast drop, but a slower slope may in theory slip through.
3. We have a minimum discharge volts setpoint, that will stop a discharge if the target voltage is detected, so on a questionable pack, you can set this high, and ass the cycling progresses, and the pack gets better, it can be lowered.
4. The new dynamic slope detect system works in parallel, and it derives a running average of the last 50 data points, and a short term average of the last 4 datapoints.
this presents a look at the last 4 datapoints as they relate to the long term running average, when in the normal operation zone, the two averages will show the same value, but on the more rapid drop out, the short term average will reflect the faster discharge slope.
The sample timing is increased to give more samples so the slope angle can be more accurately determined.
So with your eyes as a drop out test, dont blink.
With the charger in control, we have 4 ways to assure we see and stop the discharge, so the cycling procedure is much safer.
The idea is to let the thing run 3 cycles unattended by humans, and we go back and see the results when it is finished.
Of course as with the charge, we show the reason for stopping the discharge, so one can better understand what is happening. |
Insight pack end of discharge full charge #1
|  | | | Discharge and charge old insight pack |
This old pack was soak charged until the voltage stuck at 168V. Then it was discharged so we could observe the cell dropout detection, and then it was recharged with full 30 minute plateau detect. The areas of interest are plotted on separate graphs to show the details |
getting a better overview
|  | | | Draft version of data logger |
After playing with the data logger from parallax, I decided we should have a more user friendly system that would not only save the data for analysis, but show the data in real time while the charger was running. This screen grab of the program is not the final version, as it does not have the file saving set up yet, but shows how the basic system would look. The graphs are dynamic, so things like scrolling through a long data stream, changing scales, and many other cool things are built in.
We have added the capability of the PC based program to change any variable, read the actual variables, start and stop a charge, and control the charge from the PC. This allows the much more capable PC based program to do diagnostic as well as data logging and graphing the charge, as well as comparing previous charges to see how the charge profile changes with successive charges.
|
Drop out test needs more work
|  | | | Second cycle old insight pack |
The second charge of the Insight pack showed increased capacity, and the cells held up longer on the bottom end, but it also showed that our simple drop out detection system was not good enough, so we must change the way we detect the drop out.
The present thinking is that we will look at the rate of change during the linear part of the discharge, and then look for that rate of change to shorten to a percentage of that rate, as a drop out detect, this will allow us to detect the cell(s) as they begin the drop, rather than looking for a specific number on voltage drop / unit of time.
The 30 minute topping detect could also have been longer as the pack still seems to be rising when the charge was terminated |
Datalogging Labview style
|  | | | Test run with labview data logger |
I have an Appointment to meet with a Labview expert on Sunday, and I hope to come home with a runtime version of my Labview datalogger that you can download and try.
The screen shots above are where I am today, and I expect it will look quite different after my Sunday session.
This graph shows the end phase of a discharge followed by a charge then the second discharge and begining of the following charge, The chart can be zoomed to a region of interest, you can scroll around, change the scales, all while logging the data to disc and displaying the current charge.
Software also allows upload of all variables, and can permanently write new values to all eeprom locations as well as read them individually.
The charger can be started and stopped from here as well.
|
| |
|
